24-hour YouTube launch report
“I Created the Ultimate Coding Agent by Combining Pi, Codex, and Claude Code...”
The video closed its measured first-day window at 18,035 views, 765 likes, and 78 public comments. It added 17,983 views after the opening sample and averaged about 749 views/hour across almost exactly 24 observed hours. Distribution accelerated into the middle of the window, then cooled materially after roughly hour 12. The stored conversation was constructive but not uniformly celebratory: 23 positive, 31 neutral, 13 mixed, and 9 negative audience comments.
Overview
- Video
- youtu.be/5Qu2SkSQeBU
- Published
- 2026-07-24 00:21:35 UTC
- Report requested
- 2026-07-24 00:25:56 UTC — 4m 21s after publication
- Opening sample
- 2026-07-24 00:27:46 UTC — age 6m 12s; 52 views, 3 likes, 3 comments
- Final one-shot sample
- 2026-07-25 00:27:12 UTC — age 24h 05m 38s; 18,035 views, 765 likes, 78 comments
- Revalidation
- 2026-07-25 00:28:08 UTC — counters unchanged
- Observed interval
- 23h 59m 26s from opening to final one-shot sample
The requested reporting clock and the publication clock differ by 4m 21s. This report uses actual video age for every checkpoint rather than assuming that the request time was publication time.
Trajectory
| Checkpoint | Captured (UTC) | Actual age | Offset from target | Views | Likes | Comments | Segment views/h |
|---|---|---|---|---|---|---|---|
| Opening | Jul 24 00:27:46 | 0h 06m | — | 52 | 3 | 3 | — |
| Nearest 1h | Jul 24 01:28:59 | 1h 07m | +7m 25s | 746 | 55 | 10 | 680 |
| Nearest 3h | Jul 24 03:29:02 | 3h 07m | +7m 27s | 2,070 | 180 | 24 | 662 |
| Nearest 6h | Jul 24 06:31:05 | 6h 10m | +9m 31s | 4,702 | 349 | 46 | 867 |
| Nearest 12h | Jul 24 12:06:12 | 11h 45m | −15m 22s | 10,596 | 541 | 64 | 1,055 |
| Nearest 24h | Jul 25 00:27:12 | 24h 06m | +5m 38s | 18,035 | 765 | 78 | 602 |
Each nominal checkpoint uses the snapshot nearest that exact video age. Segment velocity is calculated from the preceding row, so intervals are not equal in length.
Acceleration and deceleration
- Stable opening: the first two measured segments ran at roughly 680 and 662 views/hour.
- Mid-window acceleration: pace rose to 867 views/hour from the 3h sample to the 6h sample, then to 1,055 views/hour from the 6h sample to the nearest 12h sample.
- Back-half cooling: the nearest-12h-to-24h segment averaged 602 views/hour, about 43% below the preceding segment. The last 6.16 observed hours averaged about 429/hour, and the last 3.06 hours about 345/hour.
- Cumulative pace peaked near midday: views per hour since publication reached about 903 at age 12h 16m, then eased to about 749 by the final checkpoint.
- Short burst, not a sustained re-acceleration: one 31-minute polling interval around age 16h showed 623 added views (roughly 1.2K/hour annualized), but the following intervals quickly cooled. Public counters can update in batches, so this should not be treated as a precise recommendation-system event.
Milestones
| Milestone | First observed at age | Captured (UTC) | Observed count |
|---|---|---|---|
| 1K views | 1h 37m | Jul 24 01:59:00 | 1,055 |
| 5K views | 6h 40m | Jul 24 07:01:06 | 5,129 |
| 10K views | 11h 14m | Jul 24 11:35:12 | 10,097 |
| 15K views | 17h 26m | Jul 24 17:47:19 | 15,006 |
| 18K views | 24h 06m | Jul 25 00:27:12 | 18,035 |
These are first-observed polling times, not exact crossing times.
Other first-observed engagement markers: 100 likes by age 2h 07m (114 observed), 500 likes by 10h 12m (506), and 750 likes by 22h 05m (753).
Audience sentiment
Sentiment was coded across 76 stored audience comments. No channel-owner comments or replies were present in the stored set, so the owner-excluded denominator remained 76. Neutral is the largest category because the thread contains many setup questions, factual replies, tool suggestions, and banter. Positive plus mixed comments accounted for 47.4%; clearly negative comments accounted for 11.8%.
This is a modest, self-selected public-comment sample, not a representative audience survey. The public video counter showed 78 comments while 76 audience comments were retained by the thread collector; sentiment shares use only the stored text that could actually be reviewed.
Feedback themes
What resonated
- Pi as a thin, composable harness: viewers liked the idea of minimal orchestration while still calling Codex and Claude subagents.
- Programmatic workflows: the subagent workflow, Effect/Svelte references, and ability to customize the setup drew the clearest enthusiasm.
- Practical tool discovery: Zed, cmux, Effect, and Pi capabilities prompted “I did not know it could do that” reactions.
Recurring questions and requests
- How to reproduce the Pi + Zed/cmux setup, including auto-launch behavior.
- How mobile or remote access should work, and whether T3 Code, OpenCode, OhMyPi, Collie, SSH/tmux, or another client is the better front end.
- Why custom find/grep extensions are needed if Pi already uses
fd/rgunder the hood. - How compaction behaves on long tasks, what Claude Agent SDK usage costs, and whether subscription quota applies.
- A request for an end-to-end live task that makes the advantage over vanilla Pi, Claude Code, Codex, or Cursor concrete.
Representative comments
“Minimal amount of orchestration tooling while still getting all of the benefits.” — @mc.guffin
“Will 100% steal these.” — @piesekfortnixa
“Showcase a live task… why is your setup better than vanilla Pi.” — @ashishhuddar
“This will consume your Claude usage based billing…” — @keithvertrees
Most actionable follow-up
A focused follow-up could start from a clean Pi install, complete one real coding task, and show exactly what each extension or subagent adds. A short cost/usage note and a “why this instead of vanilla Pi / Claude Code / Cursor” decision table would directly answer the most repeated skepticism without diluting the positive workflow story.
Benchmark context
The historical Ben Davis snapshot file does not contain prior uploads sampled near the same 1h, 3h, 6h, 12h, or 24h launch ages. The closest first observation for the immediately preceding upload was about 37.9 hours after publication, already well outside this report’s first-day window. Comparing this video’s 24-hour launch with older videos’ later or lifetime totals would be misleading, so no relative-performance claim is made.
Use this report as the first reusable age-matched baseline for future Ben Davis launches. The strongest future comparison will match publication age and, ideally, similar topic, format, length, traffic source mix, and day/time of release.
Methodology and limitations
- Source: public YouTube Data API v3 data collected through the read-only
oytc-based watcher. No credential file was read for this report. - Sampling: 51 stored snapshots from age 6m 12s through age 24h 06m 33s, generally at roughly 30-minute intervals. The requested final one-shot datapoint was captured at 24h 05m 38s; an additional revalidation 56 seconds later showed unchanged counters.
- Rates: view velocity is a delta between public counter snapshots divided by actual elapsed time. Like/view and comment/view are current public counters divided by current public views, not unique-viewer conversion rates.
- Checkpoint selection: 1h, 3h, 6h, 12h, and 24h rows use the nearest stored sample and disclose the exact age and offset. Milestone times are first-observed times.
- Comments: the collector stores top-level comments and replies embedded by the public comment-thread endpoint, capped at 500 threads. Sentiment excludes channel-owner comments/replies; none were present. Labels reflect overall tone: praise/support = positive, factual/questions = neutral, praise plus caveat or constructive skepticism = mixed, and clear dismissal/attack = negative.
- Counter behavior: public counters can lag or update in batches; short-interval spikes should not be over-interpreted.
- Unavailable public metrics: impressions, click-through rate, watch time, audience retention, revenue, and audience demographics are not exposed by this public API and are not estimated here.